Project context

This project was completed throughout the June 2021 data science continuous training session with the training organization Datascientest.com.

Goal

Focusing on English Premier League football matches datasets (2014/15-2017/18), this project aims at coming up with a process that could assist sports betting app users getting to a better-informed guess about a betting outcome. In order to progress on this goal, we attempted to address two challenges: predicting the matches outcome and comparing with bookmakers predictions, so as to come up with a betting strategy.

Models and packages

We experimented and compared the results for several classification machine learning models:

• Logistic regression
• K nearest neighbours
• SVM
• Decision tree and Boosting
• Random Forest
• XG Boost
• Voting classifier

We used the packages scikit-learn and XGBoost

Datasets

Data used in this project were collected via Kaggle and Datahub.io.
Data collected on Kaggle were used to retrieve information concerning English Premier League football matches for four seasons, from 2014/2015 to 2017/2018. This dataset contains match statistics and players statistics for the matches. We also collected bookmakers’ odds data for the same matches from the dataset linked above and available on datahub.io.

Repo content description

The current repo contains three notebooks:

• 1_bookmakerspy_data_collection.ipynb
  The first notebook aims at collecting data from the sources and making the necessary adjustments to obtain a dataset that contains match and players stats.
  
• 2_bookmakerspy_preprocessing_dataviz.ipynb
  The second notebooks aims at pre-processing the dataset by removing, grouping or adding variables for example. We also replaced each features row of the dataset obtained in the first notebook by the mean of the 3 past matches. The methodology is explained in more details throughout the notebook. An exploration process via visualization is also detailed.
• 3_bookmakerspy_modelisation.ipynb
  The third notebook contains the dimension reduction step, as well as the experimentations on various models. It also thoroughly details an approach tested in terms of betting strategy.

Members of the team

Mariella DE CROUY CHANEL
Thomas MIRAS
Landry TAYAYA

Our mentors from Datascientest.com

Thibault
Emilie
Antoine